*/
#define XLOG_MIN_ICLOGS 2
-#define XLOG_MED_ICLOGS 4
#define XLOG_MAX_ICLOGS 8
-#define XLOG_CALLBACK_SIZE 10
#define XLOG_HEADER_MAGIC_NUM 0xFEEDbabe /* Invalid cycle number */
#define XLOG_VERSION_1 1
#define XLOG_VERSION_2 2 /* Large IClogs, Log sunit */
#define XLOG_VERSION_OKBITS (XLOG_VERSION_1 | XLOG_VERSION_2)
-#define XLOG_RECORD_BSIZE (16*1024) /* eventually 32k */
+#define XLOG_MIN_RECORD_BSIZE (16*1024) /* eventually 32k */
#define XLOG_BIG_RECORD_BSIZE (32*1024) /* 32k buffers */
#define XLOG_MAX_RECORD_BSIZE (256*1024)
#define XLOG_HEADER_CYCLE_SIZE (32*1024) /* cycle data in header */
-#define XLOG_RECORD_BSHIFT 14 /* 16384 == 1 << 14 */
+#define XLOG_MIN_RECORD_BSHIFT 14 /* 16384 == 1 << 14 */
#define XLOG_BIG_RECORD_BSHIFT 15 /* 32k == 1 << 15 */
#define XLOG_MAX_RECORD_BSHIFT 18 /* 256k == 1 << 18 */
#define XLOG_BTOLSUNIT(log, b) (((b)+(log)->l_mp->m_sb.sb_logsunit-1) / \
#define XLOG_HEADER_SIZE 512
#define XLOG_REC_SHIFT(log) \
- BTOBB(1 << (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb) ? \
+ BTOBB(1 << (xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? \
XLOG_MAX_RECORD_BSHIFT : XLOG_BIG_RECORD_BSHIFT))
#define XLOG_TOTAL_REC_SHIFT(log) \
- BTOBB(XLOG_MAX_ICLOGS << (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb) ? \
+ BTOBB(XLOG_MAX_ICLOGS << (xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? \
XLOG_MAX_RECORD_BSHIFT : XLOG_BIG_RECORD_BSHIFT))
-/*
- * set lsns
- */
-#define ASSIGN_ANY_LSN_HOST(lsn,cycle,block) \
- { \
- (lsn) = ((xfs_lsn_t)(cycle)<<32)|(block); \
- }
-#define ASSIGN_ANY_LSN_DISK(lsn,cycle,block) \
- { \
- INT_SET(((uint *)&(lsn))[0], ARCH_CONVERT, (cycle)); \
- INT_SET(((uint *)&(lsn))[1], ARCH_CONVERT, (block)); \
- }
-#define ASSIGN_LSN(lsn,log) \
- ASSIGN_ANY_LSN_DISK(lsn,(log)->l_curr_cycle,(log)->l_curr_block);
-
-#define XLOG_SET(f,b) (((f) & (b)) == (b))
-
-#define GET_CYCLE(ptr, arch) \
- (INT_GET(*(uint *)(ptr), arch) == XLOG_HEADER_MAGIC_NUM ? \
- INT_GET(*((uint *)(ptr)+1), arch) : \
- INT_GET(*(uint *)(ptr), arch) \
- )
+static inline xfs_lsn_t xlog_assign_lsn(uint cycle, uint block)
+{
+ return ((xfs_lsn_t)cycle << 32) | block;
+}
-#define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
+static inline uint xlog_get_cycle(char *ptr)
+{
+ if (be32_to_cpu(*(__be32 *)ptr) == XLOG_HEADER_MAGIC_NUM)
+ return be32_to_cpu(*((__be32 *)ptr + 1));
+ else
+ return be32_to_cpu(*(__be32 *)ptr);
+}
+#define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
#ifdef __KERNEL__
*
* this has endian issues, of course.
*/
-
-#ifndef XFS_NATIVE_HOST
-#define GET_CLIENT_ID(i,arch) \
- ((i) & 0xff)
-#else
-#define GET_CLIENT_ID(i,arch) \
- ((i) >> 24)
-#endif
-
-#define GRANT_LOCK(log) mutex_spinlock(&(log)->l_grant_lock)
-#define GRANT_UNLOCK(log, s) mutex_spinunlock(&(log)->l_grant_lock, s)
-#define LOG_LOCK(log) mutex_spinlock(&(log)->l_icloglock)
-#define LOG_UNLOCK(log, s) mutex_spinunlock(&(log)->l_icloglock, s)
+static inline uint xlog_get_client_id(__be32 i)
+{
+ return be32_to_cpu(i) >> 24;
+}
#define xlog_panic(args...) cmn_err(CE_PANIC, ## args)
#define xlog_exit(args...) cmn_err(CE_PANIC, ## args)
#define XLOG_WAS_CONT_TRANS 0x08 /* Cont this trans into new region */
#define XLOG_END_TRANS 0x10 /* End a continued transaction */
#define XLOG_UNMOUNT_TRANS 0x20 /* Unmount a filesystem transaction */
-#define XLOG_SKIP_TRANS (XLOG_COMMIT_TRANS | XLOG_CONTINUE_TRANS | \
- XLOG_WAS_CONT_TRANS | XLOG_END_TRANS | \
- XLOG_UNMOUNT_TRANS)
#ifdef __KERNEL__
/*
/* Ticket reservation region accounting */
#define XLOG_TIC_LEN_MAX 15
-#define XLOG_TIC_RESET_RES(t) ((t)->t_res_num = \
- (t)->t_res_arr_sum = (t)->t_res_num_ophdrs = 0)
-#define XLOG_TIC_ADD_OPHDR(t) ((t)->t_res_num_ophdrs++)
-#define XLOG_TIC_ADD_REGION(t, len, type) \
- do { \
- if ((t)->t_res_num == XLOG_TIC_LEN_MAX) { \
- /* add to overflow and start again */ \
- (t)->t_res_o_flow += (t)->t_res_arr_sum; \
- (t)->t_res_num = 0; \
- (t)->t_res_arr_sum = 0; \
- } \
- (t)->t_res_arr[(t)->t_res_num].r_len = (len); \
- (t)->t_res_arr[(t)->t_res_num].r_type = (type); \
- (t)->t_res_arr_sum += (len); \
- (t)->t_res_num++; \
- } while (0)
/*
* Reservation region
} xlog_res_t;
typedef struct xlog_ticket {
- sv_t t_sema; /* sleep on this semaphore : 20 */
- struct xlog_ticket *t_next; /* :4|8 */
+ sv_t t_wait; /* ticket wait queue : 20 */
+ struct xlog_ticket *t_next; /* :4|8 */
struct xlog_ticket *t_prev; /* :4|8 */
xlog_tid_t t_tid; /* transaction identifier : 4 */
int t_curr_res; /* current reservation in bytes : 4 */
typedef struct xlog_op_header {
- xlog_tid_t oh_tid; /* transaction id of operation : 4 b */
- int oh_len; /* bytes in data region : 4 b */
- __uint8_t oh_clientid; /* who sent me this : 1 b */
- __uint8_t oh_flags; /* : 1 b */
- ushort oh_res2; /* 32 bit align : 2 b */
+ __be32 oh_tid; /* transaction id of operation : 4 b */
+ __be32 oh_len; /* bytes in data region : 4 b */
+ __u8 oh_clientid; /* who sent me this : 1 b */
+ __u8 oh_flags; /* : 1 b */
+ __u16 oh_res2; /* 32 bit align : 2 b */
} xlog_op_header_t;
#endif
typedef struct xlog_rec_header {
- uint h_magicno; /* log record (LR) identifier : 4 */
- uint h_cycle; /* write cycle of log : 4 */
- int h_version; /* LR version : 4 */
- int h_len; /* len in bytes; should be 64-bit aligned: 4 */
- xfs_lsn_t h_lsn; /* lsn of this LR : 8 */
- xfs_lsn_t h_tail_lsn; /* lsn of 1st LR w/ buffers not committed: 8 */
- uint h_chksum; /* may not be used; non-zero if used : 4 */
- int h_prev_block; /* block number to previous LR : 4 */
- int h_num_logops; /* number of log operations in this LR : 4 */
- uint h_cycle_data[XLOG_HEADER_CYCLE_SIZE / BBSIZE];
+ __be32 h_magicno; /* log record (LR) identifier : 4 */
+ __be32 h_cycle; /* write cycle of log : 4 */
+ __be32 h_version; /* LR version : 4 */
+ __be32 h_len; /* len in bytes; should be 64-bit aligned: 4 */
+ __be64 h_lsn; /* lsn of this LR : 8 */
+ __be64 h_tail_lsn; /* lsn of 1st LR w/ buffers not committed: 8 */
+ __be32 h_chksum; /* may not be used; non-zero if used : 4 */
+ __be32 h_prev_block; /* block number to previous LR : 4 */
+ __be32 h_num_logops; /* number of log operations in this LR : 4 */
+ __be32 h_cycle_data[XLOG_HEADER_CYCLE_SIZE / BBSIZE];
/* new fields */
- int h_fmt; /* format of log record : 4 */
- uuid_t h_fs_uuid; /* uuid of FS : 16 */
- int h_size; /* iclog size : 4 */
+ __be32 h_fmt; /* format of log record : 4 */
+ uuid_t h_fs_uuid; /* uuid of FS : 16 */
+ __be32 h_size; /* iclog size : 4 */
} xlog_rec_header_t;
typedef struct xlog_rec_ext_header {
- uint xh_cycle; /* write cycle of log : 4 */
- uint xh_cycle_data[XLOG_HEADER_CYCLE_SIZE / BBSIZE]; /* : 256 */
+ __be32 xh_cycle; /* write cycle of log : 4 */
+ __be32 xh_cycle_data[XLOG_HEADER_CYCLE_SIZE / BBSIZE]; /* : 256 */
} xlog_rec_ext_header_t;
#ifdef __KERNEL__
* xlog_rec_header_t into the reserved space.
* - ic_data follows, so a write to disk can start at the beginning of
* the iclog.
- * - ic_forcesema is used to implement synchronous forcing of the iclog to disk.
+ * - ic_forcewait is used to implement synchronous forcing of the iclog to disk.
* - ic_next is the pointer to the next iclog in the ring.
* - ic_bp is a pointer to the buffer used to write this incore log to disk.
* - ic_log is a pointer back to the global log structure.
* - ic_offset is the current number of bytes written to in this iclog.
* - ic_refcnt is bumped when someone is writing to the log.
* - ic_state is the state of the iclog.
+ *
+ * Because of cacheline contention on large machines, we need to separate
+ * various resources onto different cachelines. To start with, make the
+ * structure cacheline aligned. The following fields can be contended on
+ * by independent processes:
+ *
+ * - ic_callback_*
+ * - ic_refcnt
+ * - fields protected by the global l_icloglock
+ *
+ * so we need to ensure that these fields are located in separate cachelines.
+ * We'll put all the read-only and l_icloglock fields in the first cacheline,
+ * and move everything else out to subsequent cachelines.
*/
typedef struct xlog_iclog_fields {
- sv_t ic_forcesema;
- sv_t ic_writesema;
+ sv_t ic_force_wait;
+ sv_t ic_write_wait;
struct xlog_in_core *ic_next;
struct xlog_in_core *ic_prev;
struct xfs_buf *ic_bp;
struct log *ic_log;
- xfs_log_callback_t *ic_callback;
- xfs_log_callback_t **ic_callback_tail;
-#ifdef XFS_LOG_TRACE
- struct ktrace *ic_trace;
-#endif
int ic_size;
int ic_offset;
- int ic_refcnt;
int ic_bwritecnt;
ushort_t ic_state;
char *ic_datap; /* pointer to iclog data */
+#ifdef XFS_LOG_TRACE
+ struct ktrace *ic_trace;
+#endif
+
+ /* Callback structures need their own cacheline */
+ spinlock_t ic_callback_lock ____cacheline_aligned_in_smp;
+ xfs_log_callback_t *ic_callback;
+ xfs_log_callback_t **ic_callback_tail;
+
+ /* reference counts need their own cacheline */
+ atomic_t ic_refcnt ____cacheline_aligned_in_smp;
} xlog_iclog_fields_t;
typedef union xlog_in_core2 {
/*
* Defines to save our code from this glop.
*/
-#define ic_forcesema hic_fields.ic_forcesema
-#define ic_writesema hic_fields.ic_writesema
+#define ic_force_wait hic_fields.ic_force_wait
+#define ic_write_wait hic_fields.ic_write_wait
#define ic_next hic_fields.ic_next
#define ic_prev hic_fields.ic_prev
#define ic_bp hic_fields.ic_bp
#define ic_log hic_fields.ic_log
#define ic_callback hic_fields.ic_callback
+#define ic_callback_lock hic_fields.ic_callback_lock
#define ic_callback_tail hic_fields.ic_callback_tail
#define ic_trace hic_fields.ic_trace
#define ic_size hic_fields.ic_size
* that round off problems won't occur when releasing partial reservations.
*/
typedef struct log {
- /* The following block of fields are changed while holding icloglock */
- sema_t l_flushsema; /* iclog flushing semaphore */
- int l_flushcnt; /* # of procs waiting on this
- * sema */
- int l_ticket_cnt; /* free ticket count */
- int l_ticket_tcnt; /* total ticket count */
- int l_covered_state;/* state of "covering disk
- * log entries" */
- xlog_ticket_t *l_freelist; /* free list of tickets */
- xlog_ticket_t *l_unmount_free;/* kmem_free these addresses */
- xlog_ticket_t *l_tail; /* free list of tickets */
- xlog_in_core_t *l_iclog; /* head log queue */
- lock_t l_icloglock; /* grab to change iclog state */
- xfs_lsn_t l_tail_lsn; /* lsn of 1st LR with unflushed
- * buffers */
- xfs_lsn_t l_last_sync_lsn;/* lsn of last LR on disk */
+ /* The following fields don't need locking */
struct xfs_mount *l_mp; /* mount point */
struct xfs_buf *l_xbuf; /* extra buffer for log
* wrapping */
struct xfs_buftarg *l_targ; /* buftarg of log */
+ uint l_flags;
+ uint l_quotaoffs_flag; /* XFS_DQ_*, for QUOTAOFFs */
+ struct xfs_buf_cancel **l_buf_cancel_table;
+ int l_iclog_hsize; /* size of iclog header */
+ int l_iclog_heads; /* # of iclog header sectors */
+ uint l_sectbb_log; /* log2 of sector size in BBs */
+ uint l_sectbb_mask; /* sector size (in BBs)
+ * alignment mask */
+ int l_iclog_size; /* size of log in bytes */
+ int l_iclog_size_log; /* log power size of log */
+ int l_iclog_bufs; /* number of iclog buffers */
xfs_daddr_t l_logBBstart; /* start block of log */
int l_logsize; /* size of log in bytes */
int l_logBBsize; /* size of log in BB chunks */
+
+ /* The following block of fields are changed while holding icloglock */
+ sv_t l_flush_wait ____cacheline_aligned_in_smp;
+ /* waiting for iclog flush */
+ int l_covered_state;/* state of "covering disk
+ * log entries" */
+ xlog_in_core_t *l_iclog; /* head log queue */
+ spinlock_t l_icloglock; /* grab to change iclog state */
+ xfs_lsn_t l_tail_lsn; /* lsn of 1st LR with unflushed
+ * buffers */
+ xfs_lsn_t l_last_sync_lsn;/* lsn of last LR on disk */
int l_curr_cycle; /* Cycle number of log writes */
int l_prev_cycle; /* Cycle number before last
* block increment */
int l_curr_block; /* current logical log block */
int l_prev_block; /* previous logical log block */
- int l_iclog_size; /* size of log in bytes */
- int l_iclog_size_log; /* log power size of log */
- int l_iclog_bufs; /* number of iclog buffers */
-
- /* The following field are used for debugging; need to hold icloglock */
- char *l_iclog_bak[XLOG_MAX_ICLOGS];
/* The following block of fields are changed while holding grant_lock */
- lock_t l_grant_lock;
+ spinlock_t l_grant_lock ____cacheline_aligned_in_smp;
xlog_ticket_t *l_reserve_headq;
xlog_ticket_t *l_write_headq;
int l_grant_reserve_cycle;
int l_grant_write_cycle;
int l_grant_write_bytes;
- /* The following fields don't need locking */
#ifdef XFS_LOG_TRACE
struct ktrace *l_trace;
struct ktrace *l_grant_trace;
#endif
- uint l_flags;
- uint l_quotaoffs_flag; /* XFS_DQ_*, for QUOTAOFFs */
- struct xfs_buf_cancel **l_buf_cancel_table;
- int l_iclog_hsize; /* size of iclog header */
- int l_iclog_heads; /* # of iclog header sectors */
- uint l_sectbb_log; /* log2 of sector size in BBs */
- uint l_sectbb_mask; /* sector size (in BBs)
- * alignment mask */
+
+ /* The following field are used for debugging; need to hold icloglock */
+#ifdef DEBUG
+ char *l_iclog_bak[XLOG_MAX_ICLOGS];
+#endif
+
} xlog_t;
#define XLOG_FORCED_SHUTDOWN(log) ((log)->l_flags & XLOG_IO_ERROR)
xfs_daddr_t *head_blk,
xfs_daddr_t *tail_blk);
extern int xlog_recover(xlog_t *log);
-extern int xlog_recover_finish(xlog_t *log, int mfsi_flags);
+extern int xlog_recover_finish(xlog_t *log);
extern void xlog_pack_data(xlog_t *log, xlog_in_core_t *iclog, int);
extern void xlog_recover_process_iunlinks(xlog_t *log);
extern void xlog_put_bp(struct xfs_buf *);
extern int xlog_bread(xlog_t *, xfs_daddr_t, int, struct xfs_buf *);
+extern kmem_zone_t *xfs_log_ticket_zone;
+
/* iclog tracing */
#define XLOG_TRACE_GRAB_FLUSH 1
#define XLOG_TRACE_REL_FLUSH 2
#define XLOG_TRACE_SLEEP_FLUSH 3
#define XLOG_TRACE_WAKE_FLUSH 4
+/*
+ * Unmount record type is used as a pseudo transaction type for the ticket.
+ * It's value must be outside the range of XFS_TRANS_* values.
+ */
+#define XLOG_UNMOUNT_REC_TYPE (-1U)
+
#endif /* __KERNEL__ */
#endif /* __XFS_LOG_PRIV_H__ */